Institutions: The computing power industry chain presents relatively certain investment opportunities

The General Office of the People’s Government of Guangzhou City issued the Implementation Plan for Promoting High-Quality Development of the Artificial Intelligence Industry in Guangzhou. It proposed to strengthen the layout of intelligent computing power. In accordance with national and provincial policy requirements, it standardizes the hierarchical development planning layout of “city data centers + park computing power centers” and optimizes the construction standards for computing power projects. Support for market-driven construction of intelligent computing power infrastructure is encouraged, and a system of urban intelligent computing centers and distributed edge computing centers will be built according to the overall layout of “city + edge” to meet the needs of ultra-low latency business scenarios such as intelligent connected vehicles, finance, and virtual reality.

Guoxin Securities believes that 1) this round of price increases is the result of the resonance of three factors: demand explosion, supply constraints, and business model upgrades. On the demand side, the large-scale implementation of AIAgent has driven Token consumption to grow exponentially, with inference computing power replacing training computing power as the core engine; on the supply side, the global supply chain for high-end AI chips remains tight, leading to rising hardware costs. 2) The wave of price increases is driving industry prosperity to be transmitted across the entire industry chain. In the upstream hardware sector, prices of server CPUs, AI chips, and storage devices are rising; domestic chips are entering the stage of large-scale deployment, gradually matching international products in performance parameters, but there is still a gap in software ecology and high-end computing power. In the midstream computing service sector, the advantages of leading manufacturers’ full-stack layouts are being strengthened. 3) In the medium to long term, the AI computing power industry will evolve along four major directions: inference priority, full-stack competition, domestic substitution, and green intensification. Inference computing power will become the industry’s core focus, with low latency, edge-side, and high-density infrastructure becoming construction priorities; full-stack technology synergy will become a core barrier for manufacturers; the high-price cycle of computing power will be sustained; and infrastructure will upgrade towards greenness and intensification.

Bohai Securities believes that by 2026, the artificial intelligence industry is expected to officially enter the first year of AIAgent development. Currently, the AI application end continues to experience multiple catalysts, and the industry’s commercialization inflection point is expected to arrive ahead of schedule. With the dual drivers of accelerated model iteration and the application promotion by major internet companies, the computing power industry chain presents relatively certain investment opportunities.